Weak learners

• Models that perform only slightly better than random guessing are called weak learners (Bishop 2007).

• Weak learners low predictive accuracy may be due, to the predictor having high bias or high variance.

Trees may be weak learners

• May be sensitive to small changes in training data that lead to very different tree structure.
• High sensitivity implies predictions are highly variable
• They are greedy algorithms making locally optimal decisions at each node without considering the global optimal solution.
  • This can lead to suboptimal splits and ultimately a weaker predictive performance.

There’s room for improvement

• In many situations trees may be a good option (e.g. for simplicity and interpretability)

• But there are issues that, if solved, may improve performance.

• It is to be noted, too, that these problems are not unique to trees.

  • Other simple models such as linear regression may be considered as weakl learners in may situations.

The bias-variance trade-off

• When we try to improve weak learners we need to deal with the bias-variance trade-off.

The bias-variance trade-off cheatsheet

How to deal with such trade-off

• How can a model be made less variable or less biased without this increasing its bias or variance?

• There are distinct appraches to deal with this problem

  • Regularization,
  • Feature engineering,
  • Model selection
  • Ensemble learning

Ensemble learners

• Ensemble learning takes a distinct based on “the wisdom of crowds”.

• Predictors, also called, ensembles are built by fitting repeated (weak learners) models on the same data and combining them to form a single result.

• As a general rule, ensemble learners tend to improve the results obtained with the weak learners they are made of.

Ensemble methods

• If we rely on how they deal with the bias-variance trade-off we can consider distinct groups of ensemble methods:

  • Bagging

  • Boosting

  • Hybrid learners

Emnsemble methods cheatsheet

Bagging

• Bagging, Random forests and Random patches yield a smaller variance than simple trees by building multiple trees based on aggregating trees built on a subset of
  • individual (1),
  • features (2) or
  • both (3).

Boosting & Stacking

• Boosting or Stacking combine distinct predictors to yield a model with an increasingly smaller error, and so reduce the bias.

• They differ on if do the combination

  • sequentially (1) or

  • using a meta-model (2) .

Hybrid Techniques

• Hybrid techniques combine approaches in order to deal with both variance and bias.

• The approach should be clear from their name:

  • Gradient Boosted Trees with Bagging

  • Stacked bagging

Bagging: bootstrap aggregation

• Decision trees suffer from high variance when compared with other methods such as linear regression, especially when \(n/p\) is moderately large.

• Given that this is intrinsic to trees, Breiman (1996) sugested to build multiple trees derived from the same dataset and, somehow, average them.

Averaging decreases variance

• Bagging relies, informally, on the idea that:

  • given \(X\sim F()\), s.t. \(Var_F(X)=\sigma^2\),
  • given a s.r.s. \(X_1, ..., X_n\) from \(F\) then
  • if \(\overline{X}=\frac{1}{N}\sum_{i=1}^n X_i\) then \(var_F(\overline{X})=\sigma^2/n\).

• That is, relying on the sample mean instead of on simple observations, decreases variance by a factor of \(n\).

• BTW this idea is still (approximately) valid for more general statistics where the CLT applies.

The bootstrap

• Bootstrap methods were introduced by Efron (1979) to estimate the standard error of a statistic.
• The success of the idea lied in that the procedure was presented as automatic, that is:
  • instead of having to do complex calculations,
  • it allowed to approximate them using computer simulation.
• Some called it the end of mathematical statistics.

Bootstrap Applications

• The bootstrap has been applied to almost any problem in Statistics.

  • Computing standard errors,
  • Bias estimation and adjustment,
  • Confidence intervals,
  • Significance tests, …

• We illustrate it with the simplest case: estimating the standard error (that is the square root of the variance) of an estimator.

A point estimation problem

• Assume we want to estimate some parameter \(\theta\), that can be expressed as \(\theta (F)\), where \(F\) is the distribution function of each \(X_i\) in \((X_1,X_2,...,X_n)\).

• For example:

Plug-in estimators

• To estimate \(\theta(F)\) we often rely on plug-in estimators: \(\hat{\theta}=\theta (F_n)\)

Precision of an estimate

• An key point, when computing an estimator \(\hat \theta\) of a parameter \(\theta\), is how precise is \(\hat \theta\) as an estimator of \(\theta\)?

  • With the sample mean, \(\overline{X}\), the standard error estimation is immediate because the variance of the estimator is known: \(\sigma_\overline{X}=\frac{\sigma (X)}{\sqrt{n}}\)

  • So, a natural estimator of the standard error of \(\overline{X}\) is: \(\hat\sigma_\overline{X}=\frac{\hat{\sigma}(X)}{\sqrt{n}}\)

Standard error estimation

• Thus, a way to obtain a standard error estimator \(\widehat{\sigma}_{\widehat{\theta}}\) of an estimator \(\widehat{\theta}\) consists on replacing \(F\) with \(F_n\) in the ``population’’ standard error expression of \(\hat \theta\), \(\displaystyle{\sigma_{\hat \theta}= \sigma_{\hat \theta}(F)}\), whenever it is known.
• In a schematic form: \[ \sigma_{\hat \theta}= \sigma_{\hat \theta}(F) \Longrightarrow \sigma_{\hat \theta}(F_n)= \widehat{\sigma}_{\hat \theta}. \] That is, the process consists of “plugging-in” \(F_n\) in the (known) functional form, \(\sigma_{\hat \theta}(F)\) that defines \(\sigma_{\hat \theta}\)}.

The bootstrap (1)

• The previous approach, \(F\simeq F_n \Longrightarrow \sigma_{\hat \theta}(F) \simeq \sigma_{\hat \theta}(F_n)\) presents the obvious drawback that, when the functional form \(\sigma _{\hat{\theta}}(F)\) is unknown, it is not possible to carry out the substitution of \(F\) by \(F_n\).
• This is, for example, the case of standard error of the median or that of the correlation coefficient.

The bootstrap (2)

• The bootstrap method makes it possible to do the desired approximation: \[\hat{\sigma}_{\hat\theta} \simeq \sigma _{\hat\theta}(F_n)\] without having to to know the form of \(\sigma_{\hat\theta}(F)\).

• To do this,the bootstrap estimates, or directly approaches \(\sigma_{\hat{\theta}}(F_n)\) over the sample.

Bootstrap sampling (resampling)

• The bootstrap allows to estimate the standard error from samples of \(F_n\), that is,

• Substituting \(F_n\) by \(F\) carried out in the sampling step.

Bootstrap resampling (2)

• Here, \(\sigma_{\hat \theta}^*\) is the bootstrap standard error of \(\hat \theta\) and

• \(\hat \sigma_{\hat \theta}^*\) the bootstrap estimate of the standard error of \(\hat \theta\).

• That is, the new (re-)sampling process consists of extracting samples of size \(n\) of \(F_n\):
  \({\bf X^{*}}=(X_1^{*},X_2^{*},\dots ,X_n^{*})\) is a random sample of size \(n\) obtained with replacement from the original sample \((X_1,X_2,\dots ,X_n)\).

• Samples \({\bf X^*}\), obtained through this procedure are called bootstrap samples or re-samples.

The bootstrap distribution

• The distribution of a statistic computed from re-samples is called the bootstrap distribution,

• This distribution is usually not known.

• However the sampling process and the calculation of the statistics can be approximated using a Monte Carlo Algorithm.

Bootstrap Estimates of SE

• Main idea is that the bootstrap standard error of \(\hat\theta\), \(\sigma_B(\hat\theta)\) can be approximated by \(\hat{\sigma}_B (\hat\theta)\).

The bootstrap approximation, \(\hat{\sigma}_B(\hat\theta)\), to the bootstrap SE, \(\sigma_B(\hat\theta)\), provides an estimate of \(\sigma_{\hat\theta}(F_n)\):

\[ \hat{\sigma}_B(\hat\theta)(\simeq \sigma_B(\hat\theta)=\sigma_{\hat\theta}(F_n))\simeq\hat \sigma_{\hat\theta}(F_n). \]

Summary

From real world to bootstrap world:

Back to bagging

• Breiman (Breiman 1996) combined the ideas of:
  • Averaging provides decreased variance estimates,
  • Bootstrap provides multiple (re)samples.
• He suggested: bootstrap aggregating :
  • Take resamples from the original training dataset
  • Learn the model on each bootstrapped training set to get a prediction \(\hat f^{*b}(x)\).
  • Use the boostrap estimates to obtain improved prediction/classification.

Bagging prediction/classifier

• For regression (trees) the bagged estimate is the average prediction at \(x\) from these \(B\) trees.

• For classification (trees) the bagged classifier selects the class with the most “votes” from the \(B\) trees:

Out-Of-Bag observations

• Every time a resample is taken with replacement, some observations are omitted, due to the multiple occurring of others.

• These out-of-bag (OOB) observations can be used to build an estimate of prediction error.

Illustration of OOB EE

Source: https://www.baeldung.com/cs/random-forests-out-of-bag-error

Bagging in R (1.1)

• AmesHousing dataset on house prices in Ames, IA.

• We use libraries:

  • rpart for stratified resampling
  • ipred for bagging.

# Prepare "clean" dataset from raw data
ames <- AmesHousing::make_ames()

# Split in test/training
set.seed(123)
split <- rsample::initial_split(ames, prop = 0.7, 
                       strata = "Sale_Price")
ames_train  <- rsample::training(split)
ames_test   <- rsample::testing(split)

Bagging in R (1.2)

system.time(
ames_bag1 <- ipred::bagging(
  formula = Sale_Price ~ .,
  data = ames_train,
  nbagg = 100,  coob = TRUE,
  control = rpart::rpart.control(minsplit = 2, cp = 0)
)
)
#   user  system elapsed 
#  40.16    0.15   40.34

Bagging in R (1.3)

show(ames_bag1)
# Bagging regression trees with 100 bootstrap replications 
# 
# Call: bagging.data.frame(formula = Sale_Price ~ ., data = ames_train, 
#     nbagg = 100, coob = TRUE, control = rpart.control(minsplit = 2, 
#         cp = 0))
# 
# Out-of-bag estimate of root mean squared error:  26350.91 

Interpetability: The “achiles heel”

• Trees may have a straightforward interpretation,
  • Plotting the tree provides information about
    • which variables are important
    • how they act on the prediction
• Ensembles are less intuitive because
  • there is no consensus tree.
  • not clear which variables are most important

Variable importance

• A complementary way to interpret a tree is by quantifying how important is each feature.

• Done measuring the total reduction in loss function associated with each variable across all splits.

• This measure can be extended to an ensemble simply by adding up variable importance over all trees built.

Variable importance example

Random forests: decorrelating predictors

• Bagged trees, based on re-samples (of the same sample) tend to be highly correlated.
• To get away from this Breimann introduced Random forests, that use a “clever trick” that decorrelates trees:
  • When growing a tree from one bootstrap sample,
  • At each split use only a randomly selected subset of predictors.

Random forests

How many variables per split?

• The usual recommendation for random selection of variables at each split has been studied by simulation:
  • For regression default value is \(m=p/3\)
  • For classification default value is \(m=\sqrt{p}\).
• Alternatively the number \(m\) can be chosen using cross-validation.

Random forest algorithm

Random Forests Algorithm, from chapter 17 in (Hastie and Efron 2016)

Out-of-the box performance

• Random forests have become popular because they tend to provide very good out-of-the-box performance, that is:
  • Although they have several hyperparameters that can be tuned,
  • the default values tend to produce good results.
• Moreover, among the more popular machine learning algorithms, random forests have the least variability in their prediction accuracy when tuning (Probst, Wright, and Boulesteix 2019).

Out of the box performance

• Training a random forest model with all hyperparameters set to their default values, we get an OOB RMSE that is better than many other classifiers, with or without tuning.
• This combined with good stability and ease-of-use has made it the option of choice for many problems

Out of the box performance example

# number of features
n_features <- length(setdiff(names(ames_train), "Sale_Price"))

# train a default random forest model
ames_rf1 <- ranger(
  Sale_Price ~ ., 
  data = ames_train,
  mtry = floor(n_features / 3),
  respect.unordered.factors = "order",
  seed = 123
)

# get OOB RMSE
(default_rmse <- sqrt(ames_rf1$prediction.error))
## [1] 24859.27

Random forests in bioinformatics

• Random forests have been thoroughly used in Bioinformatics. See (Boulesteix et al. 2012).
• Bioinformatics data are often high dimensional with
  • dozens or (less often) hundreds of samples/individuals
  • thousands (or hundreds of thousands) of variables.

Application of Random forests

• Random forests provide robust classifiers for instance for
  • Distinguishing cancer from non cancer
  • Predicting tumor type in cancer of unknown origin
  • Selecting variables (SNPs) in Genome Wide Association Studies
• Some variation of Random forests are used only for variable selection

Another ensemble approach

• The idea of improving weak learners by aggregation has moved historically along two distinct lines:

  • Build many similar learners (trees) on resamples obtained from the original sample and, somehow, average their predictions.
    • This entails Bagging and Random Forests
  • Build a learner (tree) progressively, improving it at every step using weak learners, until the desired /maximal possible quality is obtained.
    • This is what Boosting is about.

Bagging vs Boosting

So what is Boosting

• Idea: create a model that is better than any of its individual components by combining their strengths and compensating for their weaknesses.

• For this, multiple weak models are trained sequentially, and each new model is trained to improve the errors made by the previous model.

• The final model is a weighted combination of all the models, and the weights are determined by the accuracy of each model.

Historical background

• Introduced by Robert Schapire (Schapire 1989) and Yoav Freund in the 1990s .
• AdaBoost algorithm became the first widely-used Boosting algorithm
• It achieved significant success in various applications, including face detection and handwriting recognition.
• Since then, several other Boosting algorithms have been developed, including: Gradient Boosting, XGBoost, and LightGBM.

Advantages of Boosting

• Boosting, like other Ensemble methods, improves the accuracy of weak learners and achieve better predictive performance than individual models.

• Boosting also reduces overfitting by improving the generalization ability of models.

• Available in many flavors,

• Can be parallelized

• Strong experience in Real world applications and industry.

Limitations of Boosting

• Can be computationally expensive, especially when dealing with large datasets and complex models.

• Can be sensitive to noisy data and outliers,

• May not work well with certain types of data distributions.

• Not so good as “out-of-the-box”: Requires careful tuning of hyperparameters to achieve optimal performance, which can be time-consuming and challenging.

Adaboost

• AdaBoost (Adaptive Boosting) combines multiple weak classifiers into a strong classifier.
• It does so by, at each iteration:
  • Train weak classifiers on the dataset and
  • Assign higher weights to misclassified samples.
• This way, subsequent classifiers focus more on samples that were previously miss-classified, and
• The accuracy of the ensemble classifier will increase.

Adaboost Architecture

Adaboost applications

• AdaBoost was sucessfully used in face recognition.
• Here, the task is to identify whether an image contains a face or not.
• AdaBoost can be used to train a classifier on a set of face images and a set of non-face images.
• The weak classifier can be a decision stump that checks whether a specific feature is present in the image or not.
• AdaBoost will then iteratively train weak classifiers and assign higher weights to misclassified samples, leading to a stronger ensemble classifier.
• The resulting classifier can then be used to detect faces in new images with high accuracy.

Adaboost has limitations

• Adaboost was a breakthrough algorithm that significantly improved the accuracy of ML models.
• However, Adaboost has its limitations.
  • It does not handle continuous variables very well.
  • Can be sensitive to noisy data and outliers,
  • May not perform well with complex datasets. Moreover,
  • Its performance can plateau, meaning it will no longer improve after a certain number of iterations.

Gradient Boosting

• Developed to overcome the limitations of Adaboost.
• Takes a different approach that can be linked with Optimization by Gradient Descent.
• Several advantages over Adaboost
  • Can handle continuous variables much better,
  • It is more robust to noisy data and outliers.
  • Can handle complex datasets and
  • Can continue to improve its accuracy even after Adaboost’s performance has “plateaued”.

Gradient boosting architechture

Relation with Gradient Descent

• Gradient Boosting can be seen as an extension of Gradient Descent, a popular optimization algorithm used to find the minimum of a function.
  • In Gradient Descent, the weights of the model are updated in the opposite direction of the gradient of the cost function.
  • In Gradient Boosting, the new model is trained on the negative gradient of the loss function, which is equivalent to minimizing the loss function in the direction of steepest descent.

Gradient Descent Variations

• Multiple extensions from Gradient Boosting.

• XGBoost

  • Optimized implementation that uses regularization to control overfitting and provide better accuracy.
  • Won many competitions.

• LightGBM

  • Relies on a technique to reduce the number of samples used in each iteration.
  • Faster training, good for large datasets.

Boosting applications

• Fraud Detection

• Image and Speech Recognition

• Anomaly Detection

• Medical Diagnosis

• Amazon’s recommendation engine

• Models that predict protein structures from amino acid sequences

• Pattern identification in fMRI brain scans.

Boosting application with R

• Many packages implement the many variations of boosting:
  • ada, adabag, mboost, gbm, xgboost
• An interesting option is to rely on the caret package which allows to run the distinct methods with a common interface.